llms.txt is a file at your site's root written specifically for large language models: a compact Markdown briefing that tells AI who this site is, what it covers, and which pages matter most.
Also known as llms.txt standard · /llms.txt
The problem it solves
Language models get lost in webpages more easily than people do: navigation chrome, popups, and decorative copy dilute attention, and context windows are finite. llms.txt (a community proposal from 2024) answers with a convention: instead of making AI hunt through the site, put a machine-friendly site summary at a known address: who this is, what’s here, where the details live.
The mnemonic: robots.txt is the door policy, the sitemap is the floor plan, llms.txt is the tour guide.
Format essentials
A solid llms.txt typically contains:
- H1: site or brand name.
- Blockquote: one or two sentences of positioning and credibility.
- Sectioned link lists: key pages grouped by topic, one explanatory line each.
- Optional extras: headline metrics, contact, language-version pointers.
The writing rules match AI answer visibility content rules generally: conclusions first, fact-dense, zero marketing filler; AI cites information, not adjectives.
Deploy or not?
Deploy, but don’t overrate it. llms.txt is a small brick in the technical foundation; the heavyweights of AI answer visibility remain content structure, structured data, and authority signals. The sensible move is bundling it into an AI-friendliness retrofit and doing it once. This site’s own llms.txt is a live example, generated automatically from the content system, always in sync.
Related entries: What is AI search · What is GEO
How is llms.txt different from robots.txt?
robots.txt governs access: which paths crawlers may fetch. llms.txt governs comprehension: a plain-language briefing that helps a model read the site correctly. One is a permission notice, the other is an orientation document; they complement rather than replace each other.
Do the major AI systems actually read llms.txt today?
Adoption is still settling. Some tools and retrieval systems read it; some major crawlers haven't committed. Think early-days sitemap: negligible cost, no downside, aligned with where the ecosystem is heading, a deploy-now-and-let-adoption-catch-up item.
What goes into a good llms.txt?
An H1 with the site name, a blockquote stating what the site is in one or two sentences, then sectioned link lists of key pages, each with a one-line description, all Markdown. The governing principle: save the model work. Who you are, why you're credible, and where to read what, inside one screen.
Where does it live? What about subdomains?
At the site root (example.com/llms.txt). Subdomains with distinct content should each carry their own, kept consistent with the main site; contradictory self-descriptions across entry points confuse the reader you're trying to help.